📦 openma-ai/open-managed-agents

Open Managed Agents: Your Own AI Control Center

The single most surprising thing about this project is that it lets you run a full, multi-agent AI platform directly on your own server, completely free of vendor lock-in. Open Managed Agents is an open-source toolkit that gives you total control over how artificial intelligence executes tasks. Instead of relying on a single cloud provider, you choose the specific models, execution environments, and security boundaries for every job. It solves the problem of dependency by allowing you to self-host the entire infrastructure, from the agent logic to the secure sandboxes where code runs. This means you own the data and the workflow, not the platform.

🆔 @hackernewsgithubprojects

📦 wuyoscar/jev-skill

Jev Skill: Teach Your AI Agent to Decide

Your coding agent can now make complex decisions about which tool to use, how to triage a bug, or which file to inspect before it writes a single line of code. Jev Skill is a collection of specialized instructions that you can drop into your coding agent. Instead of guessing, the agent follows a specific workflow to analyze the task, check the context, and pick the right action. It turns a general purpose AI into a focused expert that knows exactly when to ask you for help and when to proceed on its own. You stop micromanaging every step and let the agent handle the routine judgments.

🆔 @hackernewsgithubprojects

📦 danmcinerney/orchflows

Orchflows: Build Agent Workflows Once, Reuse Forever

Orchflows saves your entire multi-step AI process into a single editable text file that you can reuse forever. Most people describe their complex instructions to their coding assistant every single time they start a new session, but this tool changes that. It captures the exact steps, review gates, and stopping conditions into a simple markdown document. The truly surprising part is that these small saved procedures can be nested inside one another to build massive, complex operations without ever writing code. You define the work once, and the agent handles the rest. If you spend hours explaining your workflow to AI, this is the way to automate the setup.

📰 https://news.ycombinator.com/item?id=48960392

🆔 @hackernewsgithubprojects

📦 nasone32/llama.cpp-rdna3-7900xtx-opt

llama.cpp RDNA3 7900 XTX Opt: Squeeze Max Speed Out Of Your AMD GPU

llama.cpp RDNA3 7900 XTX Opt is the local AI build that finally squeezes real speed out of your AMD graphics card without needing a data center. Most people think consumer graphics cards are too slow for serious language models, but this project proves them wrong by treating your video memory like it belongs in a server. It stops pretending your card setup is a single, slow pipe and instead makes two graphics cards talk directly to each other to share the heavy lifting. This means you can run massive, complex AI models on your own desk with speeds that used to require expensive enterprise hardware.

🆔 @hackernewsgithubprojects

📦 89luca89/clampdown

The Sandbox That Locks Down AI

Clampdown is the container tool that keeps rogue AI agents from touching your secrets. It treats coding bots as untrusted guests, locking them in a hardened cage where they can only see their current project. Even if a prompt injection tries to trick the agent into stealing your API keys, it only gets a dummy token. The real credentials stay safely tucked away in a separate proxy, so the bot can call the cloud but never touch your actual bank account. If you let AI write code on your machine, this is the safety net you need to sleep well at night.

🆔 @hackernewsgithubprojects

📦 omacom/try-omarchy-windows

Run a Full Linux Desktop in a Windows Window

Try Omarchy for Windows turns your standard Windows machine into a host for a complete, high performance Linux desktop running inside a single application window. The most surprising capability is that it renders the entire operating system directly onto your physical graphics card without requiring dual boot, virtual machine software, or partition changes. It achieves this by leveraging Windows built in hypervisor to boot a preconfigured image, delivering smooth visuals and instant access to a modern interface. You simply download one small file, grant permission for a one time restart, and let the tool handle the rest. No complex setup or technical knowledge is needed.

📰 https://news.ycombinator.com/item?id=49780980

🆔 @hackernewsgithubprojects

📦 harnessrouter/starter-kit

Starter Kit

Build a working spreadsheet, a presentation, or even a playable Mario game, all driven by AI agents running behind the scenes. This repository gives you fully functional products that turn a simple chat into real, usable files, like a slide deck you can edit or a database dashboard that updates live. The coolest part is a game where an AI plays a browser platformer by making rapid decisions, not just looking at pixels. You get a ready-made backend that handles the heavy lifting, so you can focus on the app itself. It is a fantastic way to see how modern agents actually work in a product you can touch.

🆔 @hackernewsgithubprojects

📦 ghuntley/underclass

One Endpoint, Many AI Subscriptions: Underclass Explained

Pool multiple AI subscriptions into a single endpoint with underclass. This local proxy combines ChatGPT and Copilot accounts so your tools see one powerful provider. The standout feature is smart session pinning; it keeps your conversation on the same account to preserve expensive prompt caches. If a subscription hits its limit, underclass cools it down automatically until it resets. When everything is exhausted, it fails fast with the earliest reset time instead of hanging. Point any compatible client at this proxy to stop worrying about individual account quotas. It is a simple way to maximize your existing subscriptions.

📰 https://news.ycombinator.com/item?id=49774704

🆔 @hackernewsgithubprojects

📦 hacker-valley-media/interceptor

Interceptor: Give AI Your iPhone and Mac

Interceptor lets a single AI agent physically operate your signed-in browser, your native Mac applications, and even a real iPhone through one unified command line. The surprising capability is that it doesn't just look at screen pixels; it reads the internal structure of apps and websites, allowing it to click buttons, type text, and inspect data in the background while you keep working. This solves the problem of brittle automation by giving the agent a structured view of your actual digital life, from your personal accounts on the web to specific settings on your phone.

🆔 @hackernewsgithubprojects

📦 sepiablue-ai/comfyui-minimax-h3-w4a4-vsa

Making Video AI Fit on Your Laptop

A single twelve gigabyte graphics card can now generate high-resolution video with sound in just three and a half minutes. This ComfyUI plugin makes that happen by drastically shrinking the math required for the MiniMax H3 model. It converts massive weight files into a compact format and streams data to your GPU only when needed, solving the memory bottleneck that usually requires expensive enterprise hardware. You simply run a setup script to prepare your system and plug the workflow into your existing editor. No massive downloads or complex manual configuration are required, making advanced video generation accessible for home users.

🆔 @hackernewsgithubprojects

📦 zai-org/zcode

ZCode: The AI Coding Workspace

Take control of your coding workflow with an AI assistant that lives right where you work, whether that is a polished desktop app, a browser tab, or a simple terminal command. ZCode bundles its own terminal interface, web client, and the underlying agent runtime into one cohesive package, so you are not stitching together fragmented tools to get your code written. Instead of juggling separate scripts and servers, you get a unified environment where the agent can read your files, run commands, and help you iterate directly in the interface you prefer.

📰 https://news.ycombinator.com/item?id=49782006

🆔 @hackernewsgithubprojects

📦 logicrw/awesome-jev-projects

How Awesome Jev Maps the Agent Ecosystem

Every project in Awesome Jev is verified by checking the actual source code, not just the description. It is a directory for System-1 Agent architectures that helps developers find, understand, and compare real open-source tools. The site solves the problem of noise in the fast-moving AI world by requiring every entry to have identifiable implementation evidence at a specific commit. This means you can trust the data behind each project, including its license and how it actually makes decisions. It is built with a clean, tactile design and supports multiple languages, making it a reliable radar for finding the right tools for your needs.

📰 https://news.ycombinator.com/item?id=49782061

🆔 @hackernewsgithubprojects

📦 thrixel/build-world

Build World: Vibe Code 3D Games

Build World is the AI skill that turns plain English prompts into playable 3D worlds without writing code. It connects your favorite coding agent to Thrixel, a service that generates high-quality 3D assets while your agent handles the logic and interactions. The result is a full interactive experience like a submarine game or historical simulation, built from a single sentence. This workflow works across major engines like Three.js and Unity, letting you vibe out complex scenes and refine them with simple follow-up requests. It is a surprisingly easy way to go from idea to a live, shareable 3D experience in your browser.

🆔 @hackernewsgithubprojects

📦 jexp/neo4jev

neo4jev

neo4jev is the graph navigation tool that uses structured logic instead of unpredictable text models to find the perfect path through data. By tapping into a specialized decision-making interface, it navigates complex databases like Neo4j one single hop at a time. Instead of guessing the next step with conversational language models, it evaluates actual neighboring connections and target properties to calculate real mathematical probabilities for every possible direction. It even packs goal-checking questions into that very same step to keep things fast and efficient. Whether you are looking for a specific destination or mapping out hidden relationships, this tool transforms chaotic graph traversal into a smart, predictable beam search.

🆔 @hackernewsgithubprojects

📦 rsi-gear/gear

How Gear Makes AI Agents Better

Gear is the optimization framework that treats your AI agent's instructions as an evolving experiment. Instead of manually tweaking prompts until things work, Gear runs a two-step process where a meta agent tests different setups, analyzes exactly where the agent failed, and proposes better versions of the tools and rules. It’s like a fitness coach for your AI, but for logic. In one test, it boosted a smaller model’s success rate from twenty-seven to fifty-three percent by simply refining the workflow. You stop guessing what to change and start letting the data show you the best way to fix the agent, making it actually useful for real work.

🆔 @hackernewsgithubprojects

📦 featherless-ai/simple-jev

Simple Jev: How to Make AI Answer Instantly

Turn any standard language model into a lightning-fast decision engine. Simple Jev stops the computer from guessing one word at a time and instead reads the model's internal confidence scores to answer questions instantly. It turns raw data into clear choices or ratings without slow text generation. This means you can get structured answers like 'red' or 'true' in a blink, rather than waiting for a full paragraph. It’s a clever way to use existing models for fast, reliable logic checks. If you want your apps to make quick judgments on data, this is the tool to watch. It’s fast, it’s open, and it’s ready to try right now.

🆔 @hackernewsgithubprojects

📦 tcballard/build-omarchy-plugins

Build Omarchy Plugins: The AI Toolkit for Desktop Widgets

Build Omarchy Plugins is the AI agent toolkit that turns complex desktop widget development into a simple, guided workflow. It solves the headache of building for the Omarchy shell by teaching coding assistants how to handle tricky details like screen layouts, theme colors, and background processes. Instead of guessing, you get a set of smart skills that guide your AI helper through designing, writing, and testing your plugin step by step. This means you can create polished, professional desktop tools without needing to be a deep systems expert.

🆔 @hackernewsgithubprojects

📦 jiahao620/sam3d-part

SAM3D-Part: Click to Isolate Any 3D Part

SAM3D-Part is the 3D tool that lets you grab any piece of a model with just a click. You know that annoying problem where you need to separate a chair leg from the rest of the frame, but the software treats the whole thing as one giant block? This project solves that by letting you click on the part you want in a simple view. It then spits out that specific part as its own clean mesh, perfectly positioned where it belongs. It’s basically magic for designers who need to tweak or replace just one component without destroying the rest. One click, one isolated part, done.

🆔 @hackernewsgithubprojects

📦 rice-robotpi-lab/robotok-code

RoboTok: The Internet-Scale Hand Motion Engine

RoboTok acts as a massive search engine for human hand movements, scanning the internet to find videos where people demonstrate specific dexterous tasks. Instead of relying on slow, manual coding, it uses advanced math to map the three-dimensional shape of your hands against a huge database of real-world examples. This allows robots to learn complex grasps and manipulations by simply watching how humans naturally move. The project solves the tricky problem of teaching machines fine motor skills by turning scattered web footage into a clean, trainable dataset. It is a clever bridge between raw video and robotic intelligence, proving that we can learn by observation.

🆔 @hackernewsgithubprojects

📦 aycibatuhan/nervous-system-atlas

Nervous System Atlas: The 3D Brain That Teaches You

Nervous System Atlas is the browser-based tool that finally makes complex brain anatomy click. It runs entirely in your browser with no account or server, letting you explore over five hundred interactive models alongside real MRI slices. The coolest part is the lesion mode, which dims the scene to show exactly how a specific injury impacts the body, turning abstract text into a clear visual story. It even includes a quiz to test your knowledge. So, if you want to understand how the brain actually works without drowning in textbook jargon, this is the one tool you need to check out today.

🆔 @hackernewsgithubprojects